2.

LITERATURE REVIEW AND RELATED WORKS

2.1 BACKGROUND

Data mining and information retrieval are interdisciplinary fields

focused on extracting hidden patterns and retrieving relevant information from
vast datasets. Data mining involves uncovering insights using techniques like
machine learning [23], clustering, and classification. At the same time,
information retrieval focuses on efficiently retrieving data based on user
queries through methods like indexing and ranking. These fields combine
mathematics, statistics, computer science, and information science principles,
employing tools such as databases, envision, and machine learning
algorithms to manage, analyze, and envisage structured and unstructured
data, thereby facilitating scientific discovery and practical applications [24].
Information extraction (IE) is the automated conversion of
unstructured text into structured data, like relational databases, by identifying
entities (concepts and instances) and relationships [25]. It is vital for text
grading, converting web content into semantic web data, and identifying
clinical trial candidates [26]. Challenges include handling the complexity of
natural language and ensuring the quality of input data, which can contain
inaccuracies and inconsistencies [27]. Advanced natural language processing
and domain-specific knowledge are needed to tackle these challenges
effectively [27].
Naïve Bayes classifiers are simple probabilistic classifiers used in
machine learning. They leverage probability theory, making them highly
scalable and adaptable [28]. They require a small number of parameters that
scale linearly with the number of features in the dataset, making them efficient
for various applications, especially in text classification [29].
Psychological stress serves as a foundational factor contributing to
various health issues and mental disorders. Prolonged exposure to chronic
stress significantly elevates the likelihood of developing numerous health
problems, including insomnia, obesity, heart disease, and cancer, among
others [30].
 Figure 2.1: Effects of Stress on the Body

Figure 2.1 depicts the effects of stress on the bodies of human

beings. The Autonomic Nervous System (ANS) regulates many major
physiological functions, including as breathing, organ emission, circulatory
strain, and heart electrical activity [31]. The sympathetic nervous system
(SNS) and the parasympathetic nervous system (PNS) are the two divisions
that make up the autonomic nervous system (ANS). The SNS readies the
body's resources for action under stressful circumstances. . In contrast, the
PNS is responsible for relaxing the body and restoring it to a stable state [32].
In humans, the sympathetic nervous system (SNS) initiates the
stress response, which includes behavioral, physiological, and psychological
symptoms [33] [34]. The following definitions are applied to various answer
groups throughout this thesis: Psychological describes or has to do with the
mind or mental processes, not with carrying out an action [35]. Since
physiological reactions are an involuntary activity or reaction that makes up a
sizable amount of a living thing's everyday functioning, they might be complex
or impossible to observe with the unaided eye [36].

The dynamic process of persons or groups exchanging different

activities is known as social interaction. It was created by sociologist Erving
Goffman [15]. Social media interaction is defined as using an online platform
by people to develop social bonds with others who share their interests, goals,
backgrounds, or real-life connections [10].
In recent years, the pervasive use of social media platforms has created
an unprecedented opportunity to explore the intersection between technology
and mental health [38]. Psychological stress is a pervasive issue affecting
individuals globally, and its early identification is crucial for effective
intervention and support. The ubiquity of social media has transformed it into
a rich source of real-time, user-generated data that reflects individuals'
thoughts, emotions, and experiences. This data, if harnessed appropriately,
can be leveraged to detect and understand psychological stress, a prevalent
and often under diagnosed mental health condition [38].

Figure 2.2: Death from COVID-19

Figure 2.2 depicts the deaths that occurred during COVID-19.

Comparing experiences between nursing homes and hospitals, the study
sought to ascertain if end-of-life (EOL) conversations occurred and whether
patients were present during their last moments, particularly regarding COVID-
19 fatalities [37].
 The prevalence of psychological stress, encompassing anxiety and
depression, is a major global concern recognized by organizations like the
World Health Organization (WHO) due to its significant impact on public health
[39]. The COVID-19 pandemic, declared a worldwide epidemic and a public
health emergency by the WHO, has led to profound mental health and
psychological consequences on a global scale [8].
Social media serves as a digital platform where users openly express
a spectrum of emotions, offering a rich repository of data reflecting their mental
states. Stress is notably conveyed through various forms of social media, such
as textual content, emoticons, and social interactions [40]. Recognizing the
shift towards digital mental health solutions, there's a growing consensus on
leveraging platforms like social media for unobtrusive and continuous
monitoring, thereby presenting an opportunity to address mental health
challenges in innovative ways [13][41].

2.2 TRADITIONAL STRESS DETECTION METHODS

Traditional methods of diagnosing psychological stress, including

surveys and clinical interviews, are hindered by scalability issues, timeliness
constraints, and limitations in capturing real-world dynamics [42]. However,
physiological computing presents a promising approach to analyzing users'
psychological states using psychophysiological indicators [43]. This is
augmented by advancements in human-computer interaction and the
expansion of communication bandwidth through smart technology
development, which enhance the analysis capabilities [44]. Researchers can
now detect stress more effectively using physiological sensors, thus gaining a
comprehensive understanding of user experiences with technology.
Integration of new applications empowers even novice users in
stress detection. Psychologists recognize the interconnectedness of emotions
and physiology. Techniques such as heart rate monitoring, electrodermal
activity, heart rate variability, electromyography, pupil diameter measurement,
and galvanic skin response are employed for stress detection [45]. Traditional
methods, like EEG signals, engage users in the system to reduce stress. For
mental stress detection, a hysteresis process analyzes signals like the
electrocardiogram (ECG), pupil diameter (PD), and photoplethysmogram
(PPG), extracting relevant features through soft computing techniques.
Genetic algorithms select augmented features loaded into a fuzzy support
vector machine (FSVM) for stress and relaxation classification [45] [46].
Analysis of social media interactions on platforms where users share
content offers insights into stress states and user activity durations. Deep
learning machine learning approaches analyze stress levels based on
individual tweets and weekly posting patterns, reframing stress analysis within
the context of social media behaviour [47].

2.3 ADVANCED STRESS DETECTION METHODS

The emergence of social media as a ubiquitous platform for

communication provides an opportunity to develop an innovative hybrid model
that can automatically analyze user-generated content to detect signs of
psychological stress.
Machine learning and natural language processing (NLP)
advancements have opened new possibilities for automated large-scale social
media data analysis. Hybrid models, combining various techniques, have
shown promise in addressing the complexities of mental health detection [27].
Early detection of psychological stress is critical for preventing the
exacerbation of mental health conditions. Social media offers a platform for
proactive intervention by identifying signs of distress in real time [48].
Successful stress detection model development requires collaboration
between psychology, data science, and machine learning experts. Combining
domain knowledge with technical expertise enhances the model's
effectiveness [49].
Social media content is not limited to text; it includes videos and
other forms of multimedia. A hybrid model can effectively integrate various
data modalities to capture the complexity of stress expressions [50]. Social
media is characterized by its dynamic and ever-evolving nature. A hybrid
model can adapt to user behaviour and platform dynamics changes, providing
a more robust and resilient solution [38]. Combining different machine learning
and NLP techniques in a hybrid model can enhance accuracy and
generalization, improving the model's ability to detect stress across diverse
user populations and cultural contexts [51]. A hybrid model can facilitate real-
time or near-real-time analysis, enabling timely intervention and support for
individuals exhibiting signs of psychological stress [52].
The background reveals the intersection of mental health
challenges, social media's expressive nature, technological advancements,
and the need for innovative solutions [17]. The motivation lies in leveraging a
novel hybrid model to bridge the gap between traditional diagnostic
approaches and the dynamic, real-world expressions of psychological stress
on social media [10].

2.4 OTHER STRESS DETECTION METHODS

In Modern times, psychological stress is starting to pose a

serious threat to people's Well-Being [53]. Recognizing and controlling stress
before it worsens and causes more significant issues is vital. Figure 2.3 shows
four techniques for identifying psychological stress [32].

Figure 2.3: Taxonomy of Stress Detection Methods

 2.5 STRESS MANAGEMENT TECHNIQUES

Psychological reactions include a marked rise in unpleasant feelings

like anger, anxiety, irritation, and sadness [54]. Stress can be assessed using
a self-report questionnaire or an interview with a psychologist. [55]. The first
method, widely regarded as a reliable approach to measuring individual stress
levels, is extensively used. This includes tools such as the Stress Self Rating
Scale (SSRS), the Perceived Stress Scale (PSS), and, in some instances, the
Stress Response Inventory (SRI) [56]. However, these questionnaires only
provide information about the client's current anxiety or stress levels, not the
underlying stressors or the progression of stress. While these tests can be
administered periodically, they may not be effective for detecting subtle
changes that could signal the early stages of a major issue. Typically, they are
used only when the individual or those around them recognize or suspect the
severity of the situation, which, in most cases, is too late. Additionally,
questionnaires are subjective [57] and require the client's attention. Research
indicates that "individuals can experience memory lapses regarding the
emotional tone of a day in as little as 24 hours" [58], suggesting that people
may not always be aware of their true stress levels. Consequently, self-report
questionnaires can sometimes result in inaccurate stress level assessments
[57].
Psychological evaluation is critical in developing a stress detection
model from social media data. Understanding the intricacies of human
behaviour, language nuances, and emotional expressions is essential for
creating a hybrid model that accurately identifies psychological stress.

2.6 KEY ASPECTS OF PSYCHOLOGICAL EVALUATION IN THIS

CONTEXT

Incorporating principles of emotional intelligence is crucial. The

model should be trained to recognize and interpret emotional cues in social
media content, understanding the subtleties of stress-related expressions [59].
Deepening the model's psycholinguistic analysis capabilities enables it to
grasp the underlying emotions and sentiments in the text. This involves
understanding how language choices, tone, and linguistic patterns correlate
with psychological stress indicators [60]. Psychological stress often manifests
in specific contexts. The model should be equipped to contextualize social
media content, considering factors such as life events, relationships, and
recent experiences to assess stress [61] accurately.
Recognizing that individuals express stress differently is vital. The
model should be sensitive to diverse personality traits, coping mechanisms,
and communication styles, ensuring a nuanced evaluation that accounts for
individual differences [62]. Identifying changes in behavioural patterns is a key
aspect. The model should analyze shifts in posting frequency, engagement
with others, and content themes to detect deviations indicative of stress-
related changes. Cultural nuances significantly influence how stress is
expressed. A psychologically informed model should consider cultural
variations in language, expressions, and societal norms to avoid
misinterpretations and enhance cross-cultural applicability [63].
Analyzing how users interact with their social environment provides
valuable insights. The model should assess changes in social connections,
communication patterns, and engagement levels to understand the social
dynamics influencing stress [64]. Stress is a dynamic process that evolves.
The model should adopt a longitudinal perspective, considering historical data
and tracking changes in stress indicators, to comprehensively understand an
individual's mental well-being [65]. Ethical principles in psychological
evaluation are paramount. The model should respect user privacy, avoid
stigmatization, and prioritize the responsible use of mental health information
[66]. Collaboration with psychologists and mental health professionals is
crucial. Integrating domain expertise ensures the model aligns with
established psychological frameworks and ethical guidelines [67]. Establishing
a feedback mechanism for continuous improvement is essential. The model
should be designed to learn from user feedback and adapt to evolving
expressions of psychological stress on social media [68].
A psychologically informed approach to stress detection from social
media involves integrating principles of emotional intelligence, psycholinguistic
analysis, contextual understanding, and cultural sensitivity into the design of
the hybrid model [19]. By incorporating these psychological dimensions, the
model can achieve a more accurate and nuanced assessment of
psychological stress, contributing to early intervention and support.

2.7 PHYSIOLOGICAL INDICATORS

Stress can manifest through objectively measurable and observable

biomarkers using wearable physiological sensors. [70]. When sympathetic
nervous system (SNS) activity increases, it alters the body's hormonal levels
and triggers responses such as increased secretion production, elevated heart
rate, and heightened muscle activity. As a result, temperatures drop, and heart
rate variability (HRV) decreases [71, 69]. These alterations are indicators and
provide information via wearable sensors to identify stress. The following
categories apply to the physiological indicators of stress and the technology
that measures them.
 Activity of Brain: Electroencephalography (EEG), Functional Magnetic
Resonance Imaging (fMRI) [75].
 Activity of Heart: Electrocardiography (ECG) [72] [74].
 Response of Skin: Electrodermal Activity (EDA), Galvanic Skin Response
(GSR) [73] [76].
 Activity of Muscles: Electromyography (EMG) [77].
 Response of Respiratory: Respiratory rate monitors [78].
 Eye gaze, blinking, and pupil diameter (PD): infrared eye tracking systems
[79]
The most common methods, ECG and electrodermal activity, are
discussed below.
As discussed in this section, several physiological signals have been
employed in stress detection, with some proving effective in providing reliable
real-time information about individuals' stress levels. However, a significant
drawback of most methods is the requirement for additional hardware, which
can make them intrusive in everyday life. Despite efforts, such as those
highlighted in recent research [80], to develop wearable physiological
monitoring systems for simplicity, users are still required to continuously wear
this equipment, which can remain conspicuous and financially inaccessible for
some individuals.
Incorporating physiological signals along side social media data can
significantly enhance the accuracy and depth of psychological stress detection
[81]. Integrating physiological signals provides a holistic approach, capturing
both explicit expressions on social media and subtle physiological changes
indicative of stress. Here are key considerations for integrating physiological
signals into the hybrid model:
Integrating wearable devices, bio sensors, and cameras to collect
physiological signals such as heart rate variability (HRV), skin conductance,
electrodermal activity (EDA), and facial expressions can significantly enhance
stress detection models [82]. By understanding physiological changes
associated with stress, like increased heart rate and alterations in skin
conductance and facial expressions, we can implement real-time monitoring
systems that provide timely insights [75]. A hybrid model that processes and
analyzes physiological data concurrently with social media content enables
immediate stress detection. This multimodal fusion of physiological signals
and social media data offers a comprehensive understanding of an individual's
mental state by leveraging complementary information from both sources [83].
Effective feature extraction and representation of physiological signals
are crucial, transforming raw data into meaningful features for model
integration. Machine learning algorithms can then identify complex interactions
between social media content and physiological signals, improving stress
detection accuracy [19] and [50]. Ensuring user consent and privacy,
establishing individual physiological baselines, and aligning physiological
signals with contextual social media information is essential for ethical and
accurate stress interpretation [84].
Validating the model with ground truth data, such as self-reported stress
or clinically validated measures, ensures reliability. The model's dynamic
adaptability to changing physiological responses over time and
interdisciplinary collaboration with experts in physiology and biometrics further
enhance its effectiveness [85]. This integrated approach holds promise for
advancing stress detection and promoting holistic mental health monitoring by
combining explicit and implicit indicators of psychological stress.
 By combining social media data with physiological signals, the hybrid
model gains a more nuanced understanding of an individual's psychological
stress, leveraging explicit and implicit indicators. This integrated approach
holds promise for advancing the field of stress detection and promoting holistic
mental health monitoring.

2.8 BEHAVIOURAL RESPONSES

Behaviour is how individuals or groups respond in specific situations,

guided by established protocols, behavioural norms, or accepted social
practices [86]. Stress significantly impacts behaviour, often leading to
observable changes such as increased aggression or irritability, although
these changes can be challenging to quantify easily [87]. Other potential
behavioural shifts can be explored by analyzing how individuals interact with
devices to establish reliable methods for measuring and confirming their
association with stress.
Comparing the advantages of measuring behavioural responses
with physiological measurements, behavioural assessments often offer a
subtle approach that does not necessitate expensive additional equipment
[32].
Understanding and analyzing behavioural responses exhibited on
social media is crucial for a comprehensive psychological stress detection
model. Behavioural cues can offer valuable insights into an individual's
emotional state, providing complementary information to text and physiological
signals [88]. Integrating behavioural responses enhances the hybrid model's
ability to detect psychological stress accurately [89]. Here are key
considerations:
Behavioural indicators should be identified to enhance psychological
stress detection, including changes in posting frequency, language patterns,
social interactions, engagement levels, and specific emojis or expressions
[53]. Sentiment analysis of social media content can gauge emotional tone,
detecting shifts that signal distress [60]. Analyzing social interaction patterns,
such as reduced engagement or withdrawal, and monitoring activity and
posting patterns, like unusual posting times, can reflect stress-induced
behavioural changes. Network analysis can reveal alterations in social
connections, providing additional behavioural cues. Considering temporal
dynamics in behaviour, such as abrupt or gradual changes, offers insights into
the evolving nature of stress.
Integrating behavioural responses with text analysis and
physiological signals through multimodal fusion enhances the model's ability
to capture complex stress manifestations. Assessing user engagement
metrics, such as response rates to comments or messages, and analyzing
linguistic patterns beyond sentiment, including specific words or phrases
associated with stress, contribute to more accurate detection. Machine
learning algorithms can be trained to recognize deviations from baseline
behaviour, improving the model's efficacy.
Establishing a feedback mechanism to refine the model based on
user input ensures it adapts to individual variations in stress responses. Ethical
considerations, such as respecting user privacy, avoiding stigmatization, and
ensuring responsible use of behavioural information, are paramount in the
analysis process. Combining behavioural indicators with other data modalities,
this comprehensive approach holds promise for advancing stress detection
and promoting holistic mental health monitoring.
Integrating behavioural responses into the hybrid model makes it
more adept at capturing how individuals manifest psychological stress on
social media. Combining behavioural cues with other data modalities, this
comprehensive approach contributes to a more nuanced and accurate
psychological stress detection system.

2.9 CATEGORIZATION OF BEHAVIORAL MEASURES

The following categories apply to behavioural stress assessments and

the technology that supports them:
o Mobile phone usage: analyzing information related to users [18] to detect
behavioural patterns indicative of stress.
o Analysis of Speech: Voice stress analysis [6, 24] can be used to determine
stress levels by analysing variations in vocal features.
o Facial expression: employing automated facial expression analysis (AFEA)
to interpret emotional cues through facial movements.
o Body gesture: utilizing automated gesture analysis (AGA) to detect stress-
related body language and movements.
o Typing rhythm: assessing stress levels through keystroke and mouse
dynamics, which can indicate changes in typing speed and accuracy.
These methods provide insights into an individual's stress response by
monitoring various behavioural cues and interactions in real time or through
retrospective analysis.
In contemporary stress detection methods, speech analysis and
mobile phone usage are widely studied techniques. Behavioural stress
recognition estimations are less prevalent than physiological processes and
may not have been thoroughly researched enough, resulting in less precise
stress detection outcomes. However, these behavioural methods show
promise because they yield promising results and do not require additional
hardware, making them accessible and convenient for users.

2.10 INFORMATION RETRIEVAL MODELS

Information retrieval (IR) is widely used in various research fields,

such as information retrieval modelling, query processing, ranking, indexing,
and searching [90]. Document modelling, web searching, digital libraries,
classification, clustering, retrieval assessment, and many other areas have
text in structured, unstructured, or semi-structured formats [91]. Information
extraction and data mining techniques are extensively used in social network
analysis and email mining. Moreover, the information exists in general text,
XML and structured formats, emoticons, audio video source code,
applications, and web services [92]. The Survey of the International Data
Corporation (IDC) confirms that unstructured data occupies > 80% of digital
space [93]. IDC forecasts that in 2017, unstructured data will account for
79.2% of competence shipped, with an estimated income of 57.3%. IR aims
to find the most relevant information to the given query [94].
Definition 1: Information retrieval is finding material (usually
documents) of an unstructured nature (usually text) that satisfies an
information need from within large collections (usually on local computer
servers or from cyberspace) [95].
Definition 2: An information retrieval model is defined as a quadruple
[D, Q, F, R (qi, dj)] [96], where each component serves a specific role in the
retrieval process.
D is a set composed of logical views for the documents in the collection.
Q is a set of logical views for the user's information needs. Such
representations are called queries.
F is a framework for modelling document representations, queries, and
relationships.
R (qi, dj) is a ranking function that associates a real number with a query
qi Q and a document representation dj D. Such ranking defines an order
among the documents. D represents a collection comprising logical views of
documents.
F serves as a framework for modelling how documents and queries are
represented and their interrelationships.
R (qi, dj) denotes a ranking function that assigns a real number to a
query qi from Q and a document representation dj from D. This ranking
function establishes an order of relevance among the documents relative to
the query about query q.

Figure 2.4: Framework of Information Retrieval Search Cycle Process

 The information retrieval system performs the task of searching for
information from the past and the future. The steps involved in the search for
information, as seen by the naïve users, are depicted in Figure 2.4. However,
inside the IR model, before searching a set of documents, the pre-processing
task is executed on the Text (Query), which involves tokenization of text,
removing stop words such as articles (the, a, an), prepositions (on, at, im) and
pronouns (for example, a, about, above, according to, across, after,
afterwards. again, against, albeit, all, almost, alone, already, also, although,
always, among, as, at), performing stemming operation to find the base word
(stripping of suffixes and prefixes from the words, for example, ed, ing, ly, pre)
[97] [98], and finally the documents that are most closely matched with the
words of the query are retrieved. Documents are stored using a set of index
terms, as depicted in Figure 2.5(logical view) [94].
The traditional similarity Information Retrieval models are Boolean and
vector-space. Probabilistic and language models use the indexing process to
store documents and queries. The principles and characteristics of those
models are discussed in this section. Apart from this, other extended IR
models, such as set-theoretic models, algebraic models, and extended
probabilistic topic models, are grouped under the Representation and
Reasoning category, which is named a vertical taxonomy of IR models and
tools [99].
 Figure 2.5: Logical View of Documents: From Full Text to a Set of
Index Terms

In the year 1958, Luhn introduced the idea of information filtering

known as “Selective Dissemination of Information” [100]. This section
examines the traditional retrieval models that serve as the cornerstone of IR
research, focusing particularly on classic retrieval models that continue to
shape modern Information Retrieval techniques. However, various extended
IR models use statistical and probabilistic mathematical theories.
Information retrieval (IR) models are systems and algorithms
designed to find relevant information from large collections of data [101].
These models are essential in search engines, digital libraries, and other
applications where retrieving relevant information efficiently is crucial. Here
are some key IR models:

 Boolean Model

The Boolean model uses Boolean logic to match documents with

queries, representing both as sets of terms. It employs operators such as AND,
OR, and NOT to combine terms. While this approach is simple and easy to
implement, it cannot rank results by relevance [102].

 Vector Space Model (VSM)

The Vector Space Model (VSM) conceptualizes documents and

queries as vectors in a multidimensional space. Each dimension represents a
term, and the weight of each term can be determined by its frequency. The
similarity between a document and a query is measured using cosine
similarity, which allows the ranking of documents by relevance based on the
angle between the vectors [102].

 Probabilistic Model

Based on probability theory, the probabilistic model estimates the

likelihood that a given document is relevant to a query. The most common
probabilistic model is the Binary Independence Model, which uses the
presence or absence of terms to estimate relevant probabilities [102].

 Language Models

The language model approach treats each document as a probability

distribution over terms, with queries also treated as distributions. The goal is
to find the document model that best explains the query. A common approach
within this framework is the query likelihood model, which estimates the
probability of generating the query from each document [102].

 Latent Semantic Analysis

Latent Semantic Analysis (LSA) employs singular value

decomposition (SVD) to reduce the dimensionality of the term-document
matrix. This technique captures hidden patterns in the data, revealing
relationships between terms that may not be apparent initially. Documents and
queries are subsequently represented in this lower-dimensional space to
uncover similarities [102].

 BM25 (Best Matching 25)

BM25 is a probabilistic model that extends the Binary Independence

Model by incorporating term frequency, document length, and other factors
into the relevance score. It is widely used in search engines due to its
effectiveness in ranking documents [102].

 Neural Network-based Models

Neural network-based models use deep learning techniques to

capture complex patterns in the data. Models such as Convolutional Neural
Networks (CNNs) and Recurrent Neural Networks (RNNs) are used for
information retrieval tasks. Transformers, like BERT and GPT, have shown
significant improvements in understanding context and semantics in queries
and documents. Although these models often require large amounts of data
and computational resources for training, they provide state-of-the-art
performance [102].

 Knowledge-based Models

Knowledge-based models leverage structured data and ontologies

to understand and retrieve relevant information. They use knowledge graphs
to represent entities and relationships, enhancing the retrieval process with
semantic understanding. These models are particularly useful in domains
where precise understanding of terms and relationships is crucial, such as
biomedical information retrieval [102].
Each of these models has its strengths and weaknesses, and the
choice of model often depends on the specific requirements of the retrieval
task, the nature of the data, and the computational resources available. In
practice, hybrid models that combine elements from different approaches are
also used to achieve better performance [103].

2.11 SOCIAL MEDIA INTERACTIONS

A study indicates that an individual's writing style offers insights into

their emotional state [37], suggesting that their writing can change according
to their stress levels. Stress can sometimes enhance a person's writing
abilities, resulting in higher-quality work and a more diverse vocabulary.
Conversely, mood can be directly reflected in the content, particularly
spontaneous messages. Therefore, analyzing text linguistics can provide
valuable information for a stress recognition system.
The interactions and dynamics within social media platforms provide
a rich source of information for understanding an individual's psychological
well-being. Integrating social media interactions into a hybrid model enhances
the overall efficacy of psychological stress detection. Here are key
considerations for leveraging social media interactions in the model:
Various aspects of user behaviour must be analysed to detect
psychological stress through social media interactions [104]. This includes
examining the frequency of interactions such as posting, commenting, liking,
or sharing content, as sudden changes in these rates can indicate shifts in
well-being. Engagement levels within social circles should also be assessed,
as changes in the intensity of interactions with friends, family, or online
communities may reflect shifts in social support or isolation linked to stress.
Evaluating the sentiment of interactions, including comments and replies, can
reveal increased negativity or decreased positivity, indicating stress-related
changes in social dynamics [105]. Applying social network analysis helps
understand the structure and dynamics of social connections, where changes
in network centrality, connection alterations, or communication pattern shifts
may be stress-related.
Identifying indicators of social isolation, such as reduced responses
to mentions or declining group participation, provides valuable insights into
stress-related behavioural changes [106]. Considering peer comparisons can
also be informative, as excessive comparisons or negative reactions to peers'
achievements may signal stress [107]. Temporal patterns of interactions, such
as late-night or early-morning activity, can be associated with stress-related
sleep disturbances [108]. Assessing changes in communication style,
including shifts from detailed to brief or cryptic messages, can further indicate
stress. Monitoring participation in online communities or forums, where
reduced engagement or changes in contributions suggest stress-related
avoidance behaviours, is also crucial.
Integrating social media interaction data with other modalities like
text analysis, physiological signals, and behavioural responses enhances the
model's ability to capture the multifaceted nature of psychological stress [56]
[109], thereby ensuring user privacy.
Numerous tools, such as SentiStrength [110] and TensiStrength
[111], are currently available to automatically analyse text features. These
tools can assess writing performance through lexical measures or by directly
analyzing the "emotions" conveyed in the text, which is their primary purpose.
Researchers [112] and [111] have leveraged this technology to analyze online
posts and determine user stress levels based on the content.

2.12 SEMANTIC WEB AND SOCIAL ENGINEERING

In 1994. Tim Berners-Lee discovered the World Wide Web

Consortium named W3C (Web 2.0). Later, the term "Semantic Web" is derived
that extends the current web from the web of Linked documents to the web of
linked data, which is supposed to be an extension of XML that has RDFS with
an identifier (rdf: ID) for binding the RDF data model nodes together to build a
graph (triplets) [113]. The Semantic data of the web is well-organized and
linked so that machines use web data (metadata) for automation, integration,
interoperability, and reuse across a range of applications in a specific contest
that resembles an intelligent knowledge-based tool [114]. The foundational
elements for constructing the semantic web include Description Logic (DL),
ontology languages such as RDF/S and OWL, ontologies, and knowledge
bases [25]. As well stated in two notable statements by Tim Berners-Lee et.al,
"adding logic to web provides means to use rules to make inferences" and
"logic must be powerful enough to describe complex properties of objects"
Based on the given statement, the hierarchy of semantics is illustrated as
shown in figure 2.6 from weak semantics to strong semantics [115].

Figure 2.6: Hierarchy of Levels of Semantics

Controlled Vocabulary is a repository that stores a list of terms, for

example. To order the item(s) online from the ebay.com website, you must
click, browse, and select an item(s). Taxonomy [116] is a subject-controlled
tree classification that arranges the terms of controlled vocabulary into the
hierarchy to avoid ambiguity between the terms [117]. Similarly, a Thesaurus
is a structured collection of terms in a controlled vocabulary, interconnected
by conceptual relationships such as synonyms, homonyms, and hyponyms,
akin to WordNet [118]. Ontologies, resembling taxonomies, go beyond by
incorporating tightly integrated semantic relationships between terms
governed by rigorous rules for defining terms and their interconnections.
Ontologies have traditionally been linked with logical inference and are now
being leveraged for applications in the Semantic Web.
As computing becomes increasingly ubiquitous and pervasive, a
vast amount of information is generated in structured, unstructured, and semi-
structured formats [119]. Peter Mika has investigated the challenges that the
Semantic Web encounters with integrating social networks, which constitute a
complex network including databases, computational linguistics, knowledge-
based systems, and service-oriented computing [120]. Peter Mika's pioneering
Semantic Web Use Case, which received an award, introduces a novel
approach by applying semantic web techniques to retrieve datasets from
social networks.
This semantic web use case can personalize users' search
processes by integrating information through service discovery and analyzing
social networks like Facebook and others [121].
"Friend of a Friend" (FOAF) is an ontology designed for machines to
represent individuals, their activities, and their connections to others and
objects. FOAF enables decentralized groups to describe social networks
without relying on a central database. Using FOAF, you can create a profile
with personal information, such as name, interests, and connections to others.
This information can then be shared and interconnected with other FOAF
profiles, facilitating the creation of a distributed social network [122]. Swoogle
is the semantic web search ontology crawler that extracts metadata
information and then computes the relationship that exists between the
semantic web documents, which are stored in RDF or OWL [123]. The
information retrieval metrics used to analyse social network data (microblogs)
to know the behaviour of online users and their dynamic interactivity with other
users are studied in [124].